The Radio Neutrino Observatory-Greenland (RNO-G), deployed at Summit Station, Greenland, aims to detect ultra-high-energy (UHE) neutrinos. To maximize sensitivity, RNO-G operates with low trigger thresholds, leading to data dominated by background noise, including thermal and anthropogenic sources. A potential additional background source is associated with cosmic ray signals coming from cosmic ray-induced air-showers and further sub-showers produced in the ice, which can mimic neutrino signals. Understanding these events is crucial for improving event classification in future neutrino searches.

To address this, two parallel approaches are being developed within the collaboration. The primary method employs a linear discriminant analysis, while an exploratory approach, which this presentation is focused on, utilizes a three-stage event filtering scheme. This scheme sequentially applies a cut-based thermal noise filter, followed by machine learning classifiers—a boosted decision tree (BDT) and a convolutional neural network (CNN)—trained on both simulated cosmic-ray candidates and real background-dominated data. The method effectively rejects background while preserving high signal efficiency.

The presentation showcases this powerful machine learning based analysis, highlighting its performance in distinguishing deep cosmic-ray candidates from other types of impulsive background. These results will inform future RNO-G neutrino searches, enhancing its capability to isolate astrophysical neutrino events.